Projects

Overview: AgentLens is a trusted AI agent marketplace and workspace that connects users with specialized agents, enabling them to discover, compare, and rent agents for various tasks. The platform follows a trust-first approach, ensuring that sellers provide the specialized agents while AgentLens handles the governance and execution boundaries. Key features: - **Marketplace and Decision UI**: Users can browse, compare, and inspect trust evidence to understand how agents can be used. - **Rental-only Smart Contracts**: Time-limited access to agents, reviews, audit records, appeals, and reputation primitives. - **Audit Reference Implementation**: Six-dimensional evaluation, evidence persistence, attestation adapters, and ZK verification components. - **Agent Contract v3 and Wire v1**: Tasks, streaming events, tools, checkpoints, cancellation, resume, artifacts, and structured failures. - **Provider Contracts**: Normalized model, search, research bundle, source map, and conformance records. - **Runtime Contracts**: Immutable packages, capability grants, and R0-R4 execution-plane descriptions. Tech stack: TypeScript Why it matters: AgentLens is a trusted platform that connects users with specialized agents, making it easier for them to find and use the right agent for their tasks. The platform's trust-first approach ensures that sellers provide the specialized agents while AgentLens handles the governance and execution boundaries, making it a valuable resource for anyone looking to transform their expertise into a digital asset.

### Overview: Jarvis is a Python-based AI assistant designed to perform various tasks such as opening web pages, playing music, setting alarms, and more. It utilizes asynchronous command execution and speech recognition to provide a conversational experience. ### Key Features: - **Speech Recognition and Command Execution**: Jarvis can recognize and execute commands through voice commands. - **Web Page Opening**: Jarvis can open web pages by simply saying "Jarvis open [page name]". - **Music Control**: Jarvis can play music in YouTube by saying "Jarvis play [song name]". - **Volume Control**: Jarvis can adjust the volume of the speakers, both in terms of increasing and decreasing, as well as setting the maximum volume or turning it off. - **LibreOffice Applications**: Jarvis can open LibreOffice applications like Calc, Writer, and Impress. - **Internet Information Search**: Jarvis can search the internet for information by saying "Jarvis tells me about [topic]". - **Weather Forecast**: Jarvis can provide weather information for a given location. - **Time and Date**: Jarvis can tell the current time and date. - **Alarm Setting**: Jarvis can set an alarm. - **Internet Speed Check**: Jarvis can check the internet speed. - **Internet Connectivity Check**: Jarvis can confirm the internet connection. - **Daily News**: Jarvis can tell the daily news. - **Word Spelling**: Jarvis can spell a word by saying "Jarvis spell me the word [word]". - **Reminders**: Jarvis can create reminders. - **Linux Applications**: Jarvis can open Linux applications like Bash and Firefox.

Overview: FaceTracker is a library for real-time deformable face tracking developed using C++ and OpenCV 2, authored by Jason Saragih and maintained by Kyle McDonald. It is designed for applications requiring face tracking in real-time scenarios. Key features: - Real-time deformable face tracking - Compatible with OpenCV 2 - Supports various platforms including Android, iOS, and native C++ applications Tech stack: C++ and OpenCV 2 Why it matters: FaceTracker is useful for applications requiring real-time face tracking, such as security systems, surveillance, and interactive applications where accurate face detection and tracking are crucial. Repository: kylemcdonald/FaceTracker GitHub stars: 1014 GitHub description: Real time deformable face tracking in C++ with OpenCV 3. README excerpt: This repository is no longer maintained, due to challenges of upgrading to OpenCV 4. FaceTracker is a library for deformable face tracking written in C++ using OpenCV 2, authored by Jason Saragih and maintained by Kyle McDonald. It is designed for applications requiring face tracking in real-time scenarios.

### Overview: Causal Forcing and Causal Forcing++ are innovative methods for generating high-quality real-time interactive video using causal diffusion distillation. These techniques aim to improve the visual quality and motion dynamics of video generation, offering a more efficient and accurate initialization compared to traditional methods. ### Key Features: - **Causal ODE or Causal Consistency Distillation**: Drive asymmetric DMD as a theoretically correct initialization for real-time interactive video generation. - **Chunk-wise and Frame-wise Models**: Support both chunk-wise and frame-wise models, with the latter natively unifying T2V and I2V. - **Causal Forcing++**: Replaces ODE with causal Consistency Distillation to eliminate ODE data curation, releasing the first 1-step/2-step frame-wise models. ### Tech Stack: Python ### Why it Matters: - Significantly outperforms Self Forcing in both visual quality and motion dynamics, while maintaining the same training budget and inference efficiency. - Supports both chunk-wise and frame-wise models, with the latter natively unifying T2V and I2V. - Released the first 1-step/2-step frame-wise models, improving performance and efficiency. - Built on top of state-of-the-art video world models, such as DreamX-World 1.0, Matrix-Game 3.5, Astronex-World 1.0, Zing-0.5, and ZYT-World, demonstrating its effectiveness in various applications.

Overview: Sherpa is an open-source speech-to-text inference framework built with PyTorch, focusing on end-to-end models such as transducer- and CTC-based models. It provides both C++ and Python APIs for deployment. Key features: - C++ and Python APIs - Focus on deployment, using pre-trained models for speech transcription - Supports PyTorch, PyTorch Lightning, and PyTorch XLA - Other similar projects: sherpa-onnx, sherpa-ncnn Tech stack: C++ Why it matters: Sherpa simplifies the process of deploying speech-to-text models by providing pre-trained models and APIs, making it accessible for developers and researchers alike. GitHub stars: 1013

Overview: Kandev is an AI Kanban & Development Environment that allows power users to customize workflows, agents, and runtimes, review changes, and ship value. It supports multiple providers and is self-hostable with no telemetry. Key features: - Multi-agent support - Parallel task execution - Integrated workspace - Kanban task management - Agentic workflows - Automations - Sub-tasks - CLI passthrough Tech stack: Go Why it matters: Kandev empowers teams to run AI agents in a controlled and efficient manner, ensuring human oversight and trust in the deployed code. Repository: kdlbs/kandev GitHub stars: 990 GitHub description: AI Kanban & Development Environment. Orchestrate multiple agents, review changes, open PRs. Multi-provider, self-hostable, no telemetry.

Overview: A Python-based desktop voice assistant designed for executing system-level commands, integrating speech recognition and text-to-speech, and handling asynchronous user interactions. Key features: - Bilingual support (French/English) - Cross-platform compatibility (Linux, Windows) - Greeting based on the time of day - Time and date in the chosen language - System control: Restart or shut down system via voice commands - Web browsing: Open YouTube, Google, or any website - Wikipedia: Search for information and get a voice summary - Entertainment: Play music from local library or tell jokes - Productivity: Take screenshots and save notes Tech stack: Python Why it matters: This project offers a practical and efficient solution for automating daily tasks on personal computers/laptops, providing a more intuitive and user-friendly interface compared to traditional methods. Repository: kishanrajput23/Jarvis-Desktop-Voice-Assistant GitHub stars: 1007

Overview: This repository contains a collection of Jupyter Notebooks designed for Natural Language Processing (NLP) enthusiasts and developers. It serves as a comprehensive resource for learning and practicing various NLP techniques, including word embeddings, named entity recognition, text classification, sentence similarity, multilingual word embeddings, transfer learning, and more. Key features: - Word Embeddings - Named Entity Recognition - Text Classification - Sentence Similarity - Multilingual Word Embeddings - Transfer Learning Tech stack: Jupyter Notebook Why it matters: This collection of Jupyter Notebooks provides a practical and accessible way for developers to learn and apply NLP techniques, making it an invaluable resource for anyone interested in advancing their NLP skills.

Overview: This repository contains pre-trained GANgealing models for eight datasets, including the Spatial Transformers and generators. It also provides training code, scripts, and a Colab notebook for mixed reality applications, as well as a CUDA implementation for generating high-quality warping visualizations and an anti-aliased grid sampling implementation for Spatial Transformers. Key features: - Pre-trained GANgealing models for eight datasets - Training code for Distributed Data Parallel and torchrun API support - Colab notebook for mixed reality applications - CUDA implementation for generating high-quality warping visualizations - Anti-aliased grid sampling implementation for Spatial Transformers Tech stack: Python, PyTorch, CUDA 10.2/cuDNN 7.6.5 Why it matters: This repository provides a comprehensive solution for dense visual alignment using GAN-Supervised Learning, offering pre-trained models and training scripts to facilitate research and application development.

Overview: This repository contains experiments and projects exploring the capabilities of ChatGPT and the Code Interpreter plugin, showcasing what can be achieved with this powerful combination. Key features: - Code Interpreter plugin integration - Step-by-step guide on activating Code Interpreter - Limitations and tips for using the environment - Instructions for installing external Python packages Tech stack: Python, Jupyter Notebook Why it matters: This repository provides a practical demonstration of how to leverage ChatGPT and the Code Interpreter to perform various tasks, including data analysis, image conversion, and code editing, showcasing the potential of this combination.

Overview: This repository provides a simple implementation of the GPT-2 model using PyTorch, focusing on text generation. It includes a pre-trained model and detailed instructions for usage. Key features: - GPT-2 model implementation - Text generation capabilities - PyTorch framework Tech stack: PyTorch Why it matters: This implementation serves as a practical example for understanding and experimenting with GPT-2 models, particularly in the context of text generation tasks. Repository: graykode/gpt-2-Pytorch GitHub stars: 1012 GitHub description: Simple Text-Generator with OpenAI gpt-2 Pytorch Implementation

### Overview: KakaoBrain KoGPT is a state-of-the-art Korean Generative Pre-trained Transformer model designed for natural language processing tasks. This model is optimized for high performance on GPU hardware and is built on the Hugging Face platform, making it accessible and widely used in various applications. ### Key Features: - **Model Type**: KoGPT6B-ryan1.5b - **Model Size**: 6,166,502,400 parameters - **Model Layers**: 28 layers - **Model Dimensions**: - d_model: 4,096 - d_ff: 16,384 - n_heads: 16 - n_ctx: 2,048 - n_vocab: 64,512 - **Positional Encoding**: Rotary Position Embedding (RoPE) - **Hardware Requirements**: - GPU: 32GB RAM or higher - Half-precision: NVIDIA Volta, Turing, or Ampere GPUs - **Usage**: - **Prompt**: ```bash python -m kogpt --help ``` ```bash python -m kogpt ``` - **Model Inference**: ```python import torch from transformers import AutoTokenizer, AutoModelForCausalLM tokenizer = AutoTokenizer.from_pretrained( 'kakaobrain/kogpt', revision='K

### Overview: CozyClay is an open-source previs software that enables you to create scenes and videos in a browser tab. It allows you to block the set, pose characters, and author camera moves and cuts, then take the same shots to an AI video model. ### Key features: - **Scene Blocking**: Block the set and characters to create a detailed scene. - **Motion Compilation**: Compile motion and QA for collisions, foot slides, reach, and hand contact. - **Camera Shots and Cuts**: Author camera moves and cuts, ensuring smooth transitions. - **AI Video Model**: Export the scene as a greybox reference clip and prompt, which the AI video model uses to generate a video. ### Tech stack: JavaScript ### Why it matters: CozyClay simplifies the process of creating videos by allowing you to make decisions in 3D, making it cheaper and more efficient. It also eliminates the need for expensive AI video models, making it accessible to a wider audience.

### Overview: FlashRank is a lightweight and ultra-fast Python library designed to enhance your search and retrieval pipelines by adding re-ranking capabilities. It supports both Listwise and Pairwise rerankers, leveraging SoTA models like T5 and cross-encoders. ### Key Features: 1. **Ultra-Lite**: Runs on CPU without the need for Torch or Transformers, with a tiny model size of ~4MB. 2. **Super-Fast**: Reranks tasks efficiently, with time complexity dependent on the number of tokens and model depth. 3. **Cost-Conscious**: Offers low $ per invocation, with smaller package size for serverless deployments. 4. **SoTA Models**: Based on state-of-the-art cross-encoders and other models, ensuring high performance. ### Tech Stack: Python ### Why It Matters: FlashRank provides a lightweight and efficient solution for re-ranking tasks, making it ideal for developers looking to enhance their search and retrieval pipelines without compromising on performance. Repository: PrithivirajDamodaran/FlashRank Primary Language: Python GitHub Stars: 1010 GitHub Description: Lite & Super-fast re-ranking for your search & retrieval pipelines. Supports SoTA Listwise and Pairwise reranking based on LLMs and cross-encoders and more. Created by Prithivi Da, open for PRs & Collaborations.

### Overview: TinyLLaVA Factory is an open-source modular codebase for small-scale large multimodal models (LMMs), implemented in PyTorch and Hugging Face, focusing on simplicity, extensibility, and reproducibility. ### Key features: - **Model Zoo**: Includes a suite of cutting-edge models and methods. - **LLM**: Supports OpenELM, TinyLlama, StableLM, Qwen, Gemma, and Phi. - **Vision Tower**: Supports CLIP, SigLIP, Dino, and combinations of CLIP and Dino. - **Connector**: Supports MLP, Qformer, and Resampler. - **Training Recipe**: Supports frozen, fully, and partially tuning, and LoRA/QLoRA tuning. - **Installation and Requirements**: Upgrade to the latest code base. - **Get Started**: - **1. Data Preparation**: Prepare your data. - **2. Train**: Train your model. - **3. Evaluation**: Evaluate your model's performance. - **Model Zoo**: - **Trained Models**: Model performance details. - **Legacy Models**: Existing models for reference. - **Launch Demo Locally**: - **Gradio Web Demo**: Accessible via a web interface. - **CLI Inference**: Command-line inference scripts. - **Quick Inference Scripts**: Simplified inference scripts for quick testing. - **Custom Finetune**: Customize your own large multimodal models. - **Customize Your Own Large Multimodel

### Overview: 本项目提供了一个包含中国计算机视觉(CV)算法岗位的公司名单，旨在帮助求职者了解和寻找合适的CV算法岗位。求职者可以通过提交issue来补充和完善这个列表。 ### Key features: - **公司列表**: 包含了来自国内外的计算机视觉(CV)算法岗位的公司信息。 - **更新频率**: 每日更新，确保信息的时效性和准确性。 - **补充功能**: 通过提交issue，求职者可以补充或更新公司信息。 ### Tech stack: - **GitHub**: 用于存储和管理项目代码。 ### Why it matters: 这个项目为求职者提供了全面的CV算法岗位信息，帮助他们快速找到合适的职位，提高求职成功率。

### Overview: Building and evaluating AI agents is crucial for advancing the field and ensuring the development of robust and ethical AI systems. This curated library serves as a comprehensive resource for researchers, practitioners, and enthusiasts, providing a wealth of information on papers, blogs, talks, tools, and benchmarks. ### Key Features: - **Must-read Starter Set**: Essential papers and blog posts that explain the importance of evaluation in AI development. - **Model / Harness / Skill Decomposition**: Understanding the components of AI models and how they can be evaluated. - **Observability & Output Space**: Techniques for grading outputs and understanding how AI agents perform. - **Evaluation Infrastructure**: Detailed guides on datasets, scorers, and evaluation methods. - **Benchmark vs. Evaluation**: Clarification on the differences between benchmarks and evaluations. - **LLM-as-Judge & Verifiers**: Discussion on the alignment and biases of AI models and how to verify their outputs. - **Agent-Specific Evaluation**: Techniques for evaluating AI agents in various scenarios, including trajectories, tool use, multi-turn interactions, world state, and multi-agent systems. - **Safety / Adversarial Evaluation**: Methods for evaluating AI agents in a safe and adversarial environment. - **Talks, Podcasts, & Slides**: Transcribed and annotated transcripts of talks and podcasts. - **Companies & Landscape**: Insights into the AI evaluation and RL environment market. - **Notes on Provenance & Gaps**: Deep notes on the provenance and gaps in evaluation methods. - **Contributing**: Guidelines for contributing to the project. ### Tech Stack:

Overview: This repository contains curated cheatsheets for core topics in Artificial Intelligence, Machine Learning, and Deep Learning, designed to help students, developers, and researchers quickly recall important concepts and formulas. Key features: - Artificial Intelligence - Transformers and LLMs - Deep Learning - Machine Learning - Probabilities and Statistics - Algebra and Calculus Tech stack: Python, Markdown, Git, and GitHub. Why it matters: These cheatsheets provide quick access to essential information for students, developers, and researchers in the fields of artificial intelligence, machine learning, and deep learning. GitHub stars: 1008

### Overview: This repository serves as a comprehensive survey and benchmarking platform for recent open vocabulary learning methods, focusing on tasks such as open-vocabulary object detection, segmentation, and tracking. It aims to supplement the authors' survey paper published in T-PAMI 2024. ### Key Features: - Detailed results and fair comparisons for the most representative works in open vocabulary learning. - Comprehensive coverage of related domains, including foundation model tuning and open-world detection. - Aiming to provide a clear and fair comparison of different approaches in open vocabulary tasks. ### Tech Stack: Python, Git, and various machine learning libraries. ### Why it Matters: This survey is crucial for researchers and practitioners in the field of open vocabulary learning, offering a comprehensive overview of the latest methods and their performance. It serves as a valuable resource for understanding the current state of the art and identifying promising future directions.

### Overview: AutoGen UI is a web-based user interface designed to integrate with the AutoGen framework, enabling developers to quickly build UIs for multi-agent LLM applications. The project utilizes TypeScript and FastAPI for the backend and Next.js for the frontend, providing a simple yet powerful solution for automating the creation of user interfaces. ### Key Features: - **AutoGen AgentChat Interface**: The UI is built using the AutoGen AgentChat API, allowing developers to interact with predefined agent teams through a simple chat interface. - **Next.js Frontend**: The frontend is built with Next.js, providing a straightforward and efficient way to handle client-side rendering and state management. - **FastAPI Backend**: The backend is powered by FastAPI, enabling the server-side processing and handling of requests. - **Manager Class**: A simple manager class is provided to run tasks and stream results to the frontend, making it easier to manage and interact with the backend. ### Tech Stack: - TypeScript - FastAPI - Next.js ### Why It Matters: AutoGen UI simplifies the process of building UIs for multi-agent LLM applications, making it easier for developers to create interfaces without extensive coding knowledge. The integration with AutoGen AgentChat API ensures that the UI can be easily configured and managed, reducing the need for complex setup and configuration.